Spatial heterogeneity analyses identify limitations of epidemic alert systems: Monitoring influenza-like illness in France

Background. Syndromic surveillance data are typically fully aggregated in space, to be used for epidemic alert systems. However, epidemics may be spatially heterogeneous, undergoing distinct dynamics in separate regions of the surveillance area. Global alerts, triggered by spatially aggregated data, may be thus suboptimal, arriving either too late or too early for the local authorities. Methods. We develop new analyses of incidence time series to assess spatial heterogeneity of epidemics, using metapopulation theory. We quantify the relationship between local time series using Pearson correlation and use modularity to classify the local data. The surveillance area is thus divided into regions with different incidence patterns. We analyze 31 years of surveillance data on influenza-like-illness (ILI) from the French system Sentinelles. Results. Spatial heterogeneity in ILI spread was found in 19/31 seasons. Distinct epidemic regions could be identified as soon as five weeks after the nationwide alert. We found three scenarios for the role of spatial heterogeneity in ILI epidemiology. First, when the nationwide epidemic alert was triggered, 32-41% of the administrative regions were experiencing an epidemic, while the others were not. Second, the nationwide alert was timely for the whole surveillance area. However, past the alert, regions experienced distinct epidemic dynamics. Third, the epidemic dynamics was homogeneous in the entire surveillance area.